Shear measurement bias II: a fast machine learning calibration method

TL;DR

This paper introduces a fast machine learning-based shear calibration method that accurately estimates shear responses from image properties, achieving low bias and efficiency suitable for large surveys like Euclid.

Contribution

A novel supervised machine learning approach for shear calibration that is computationally efficient, reduces data requirements, and performs comparably to existing methods like Metacalibration.

Findings

Residual bias compatible with zero on simulated data

Achieves Euclid requirements for SNR > 20

Requires only ~15 CPU hours for training

Abstract

We present a new shear calibration method based on machine learning. The method estimates the individual shear responses of the objects from the combination of several measured properties on the images using supervised learning. The supervised learning uses the true individual shear responses obtained from copies of the image simulations with different shear values. On simulated GREAT3data, we obtain a residual bias after the calibration compatible with 0 and beyond Euclid requirements for a signal-to-noise ratio > 20 within ~15 CPU hours of training using only ~10^5 objects. This efficient machine-learning approach can use a smaller data set because the method avoids the contribution from shape noise. The low dimensionality of the input data also leads to simple neural network architectures. We compare it to the recently described method Metacalibration, which shows similar performances. The different methods and systematics suggest that the two methods are very good complementary methods. Our method can therefore be applied without much effort to any survey such as Euclid or the Vera C. Rubin Observatory, with fewer than a million images to simulate to learn the calibration function.